Condition monitoring of IGBT modules using online TSEPs and data-driven approach

The nonlinear operating characteristics associated with the power electronic converters result in high mechanical and thermal stresses for the power modules. These stresses collectively result in failure modes such as wear-out and sudden failures, which lead to power loss for the system. To overcome these aspects, this paper proposes a condition monitoring approach based on data-driven methods for power modules. The proposed approach is aimed at achieving early detection of power module failure mode by measuring the temperature sensitive electrical parameters (TSEPS) with a measurement and monitoring unit (MMU). To develop the proposed monitoring approach, 12 different online TSEPs of insulated gate bipolar transistor (IGBT) power module are measured with the MMU and trained with a support vector data descriptor classifier. The training process is carried out with a two-chip half bridge IGBT leg which is further operated in a single-phase full bridge inverter configuration during the testing process. Both the training and testing processes of the developed classifier are carried out for IGBT operation at multiple temperatures, and multiple faults to identify the repeatability of the TSEPs. The results of numerical simulations and experimental analysis identified 96.3% training accuracy and 2.9% error in testing accuracy at 90°C junction temperature for degradation identification. For the failure mode classification, the training accuracy is identified to be 98.6%, and 100% testing accuracy at 25°C.